In April a founder in Vilnius told me she had stopped using the word “engineer” in her bio. Not because she had pivoted. Because she didn’t know what it meant anymore.
She still wrote code. She shipped a product last quarter. But she had spent the previous month watching a non-technical co-founder build a working prototype using Claude Code and three weekends of effort. “He’s not an engineer,” she said. “But what I do now is mostly review what models write. So what am I?”
This is the conversation I keep having. Different cities, different roles, same shape. Over the past six months I have spoken with more than sixty founders, product managers, designers and engineers across Europe and the CIS for this publication. They are not, mostly, worried about losing their jobs. They are worried about something stranger. They no longer know what their jobs are for.
The data supports the unease. A Stanford study cited in early 2026 reporting found that employment for software developers aged 22 to 25 had fallen nearly 20% since 2022, a structural shift in the entry-level market. The Pragmatic Engineer’s April 2026 survey of more than 900 engineers, run by Gergely Orosz and Elin Nilsson, found that some of the most experienced “builders” report a sense of identity loss and even some grief, because they can no longer justify hands-on coding when an agent generates decent code faster than they can type. Meanwhile, Atomico’s State of European Tech 2025 reports that 31% of all European funding raised in 2025 is going to AI/ML companies. The rest of the ecosystem now builds, hires and prices in proximity to this category whether it wants to or not.
The labels we use (engineer, PM, marketer, designer) were built to describe scarce skills. When the skill becomes abundant, the label stops being a description and starts being a costume. Most people I speak with sense this. Very few have a clean answer. But the answers, when they come, group into three patterns. The patterns matter, because they suggest where the next phase of professional identity in tech will come from, and where it won’t.
Pattern one: narrowing until the model can’t follow.
The first response is hyperspecialization pushed past the point of comfort. The new titles I hear are narrow to the point of awkwardness. “PM for regulated AI products in DACH.” “Fine-tuning specialist for Polish-language enterprise LLMs.” “Designer of multi-agent workflow interfaces.”
This is the most data-supported of the three patterns. A KORE1 2026 staffing analysis found that more than 75% of AI job postings now specifically call out deep domain expertise, and that specialists with the right niche pull salaries 30% to 50% above generalists at the same level. PwC’s 2025 Global AI Jobs Barometer puts the salary uplift for engineers with AI-specific skills at up to 56%.
The people I met in Berlin in March who had pushed this furthest had something in common, and it wasn’t their CV. None of them had degrees in their stated specialty. They had invented the niche, written about it consistently for a year, and built quiet authority inside small invite-only Slacks and Discord servers, communities of 200 to 500 people where most of the substantive work in their sub-field actually happens. Their bet is structural: a model can do general work cheaply, but it cannot manufacture a five-year track record in a category that did not exist five years ago.
The risk is obvious. The niche can shrink faster than you can pivot. I watched one of these specialists go from in-demand to invisible in eight weeks last autumn, because the regulatory ground he had staked moved underneath him. Hyperspecialization buys you a moat that is also a cell.
The CIS version of this pattern looks different, and the difference is worth naming. In the Russian-speaking ecosystem the move is more often outward than inward. The anchor is geographic or regulatory. It rarely tracks a technical sub-domain. “PM for fintech compliance in Kazakhstan” travels further than “PM for regulated AI.” The 15,000 to 20,000 Russian IT specialists who relocated to Kazakhstan since 2022, and the roughly 9,000 Russian-founded businesses now registered in Belgrade, where ICT exports grew 20% year-on-year to €4.3B, have produced a generation of operators whose specialism is mostly where they work, not what they do. It is hyperspecialization by ZIP code. It works, until it doesn’t.
Pattern two: identity by witness.
The second response is more familiar. Build in public. Document the work. Post the doubts. Let an audience of strangers ratify, in real time, that you are the thing you say you are.
In the CIS this almost always lives on LinkedIn. The platform has eaten Russian-speaking tech in a way that has no real Western parallel. X never settled here. Substack is a niche. In Western Europe the pattern is more diffuse: Substacks, podcasts, semi-public Discord servers, the occasional operator newsletter. The medium varies. The mechanic is the same. If enough people watch you work, the question of what you are gets answered by witnesses rather than by definitions.
This worked, once. It has now run into a wall, and the wall is empirical. An Originality.AI study from 2025 covering hundreds of thousands of LinkedIn posts found that posts flagged as likely AI-generated averaged roughly 30% less reach and 55% less engagement than human-written ones, and that more than half of long-form LinkedIn posts that year were likely AI. The average LinkedIn post length, over the same period, has risen from well below 500 words pre-2023 to just under 1,500 today. A medium designed for thinking out loud has become a medium for sounding like thinking.
The performance version of building in public has been catastrophically commoditized by the same models people use to do their actual work. The honest version is rare and increasingly valuable. The people doing it well in 2026 have one thing in common: they are demonstrably worse at the performance than they are at the work. The unfinished sentence, the slightly off photograph, the post that explains something poorly but truthfully, these now read as authentication. A polished thread reads as a model.
The third move is harder to write about, because the people doing it do not want to be written about.
They have stopped explaining themselves. No bio updates, no thought leadership, no rebranding. They ship. They take meetings. They are referred by other people who know what they actually do. When you ask what they are, they tell you what they are working on, not what they are.
Several of the most interesting products I have seen in early 2026 (two in Tallinn, one in Lisbon, one in Tbilisi) are made by people in this category. None of them have a persona. None of them are coming up in your feed. They are doing the strange thing of letting the work be the identity, which used to be unremarkable and is now almost a stance.
It looks like stoicism or modesty. It is closer to a calculation: in a market saturated with people performing competence, real competence is once again differentiated by silence. They may be wrong. There is a credible argument that the quiet builders of 2026 will be the unknown casualties of 2027, outcompeted by louder peers who got the distribution. I do not know yet. I notice that the quiet ones, so far, are not losing.
When I press them, the answer is observational. They noticed that the energy required to maintain a professional identity online had started to exceed the energy required to do the work. So they stopped maintaining the identity. A 2024 WithDouble survey found that 53% of startup founders reported burnout; a 2025 CEREVITY analysis of 127 California tech founders found that 73% of those experiencing persistent burnout symptoms were still meeting or exceeding their business targets. The cost of the performance has stopped being theoretical.
None of the three patterns is a solution. They are coping strategies for a market still in the early phase of a structural change. Hyperspecialization is a moat that erodes. Identity-by-audience is a treadmill that AI is now running faster than its human inhabitants. The refusal works for a small number of people with strong networks and weakens for anyone without one.
But the patterns are useful for a different reason. They expose that the question “what do you do” has stopped being a useful question in tech. It was a useful question when skills were scarce and labels were honest. It is a brittle question now. Atomico recorded more than 27,000 new founders starting companies in Europe in 2025, around 60% more than in 2023 and the highest number on record. The supply of people building has gone up. The supply of stable categories to describe them has gone down. The mismatch has to land somewhere, and right now it is landing on individuals.
The people I find most clear-eyed in 2026 have stopped trying to answer “what do you do” well. They are answering a different question: what does the next year of my work need to prove, and to whom?
This is a smaller question. It is also a more honest one. It does not require an identity. It requires a sequence of moves. The identity crisis in AI-era tech ends when we stop needing labels. Finding new ones will not fix it.
The founder in Vilnius hasn’t reached that yet. She still wants a word for what she does. But she has noticed that the word she had isn’t coming back. Most of the people I talk to are at that stage. The interesting question, for this publication and for the rest of the year, is what they do next.
